How to judge a bid change without fooling yourself

Somebody changes a target or a budget on a Monday. By Friday there is a chart, and the chart becomes the verdict. It is rarely anyone's fault: the first week is the only evidence available when the question is asked.

The trouble is that both platforms publish rules for when a change can fairly be read, and the first week can fall short of them. There are four things they publish that change how a bid change should be read, and each of them has its own entry in our glossary.

If the target moved during the window, the number you are comparing against is not the number the strategy aimed at. Google reports average target CPA, the traffic-weighted average CPA that your bid strategy optimised for, and says to compare the CPA you achieved with it because it more accurately reflects what Smart Bidding is optimising towards.

Our rule follows from it: the first question about any Target CPA change is not whether CPA is above the target. It is whether CPA is above the average target for the same dates. Google's Target CPA page also recommends measuring performance over periods that have at least 30 conversions, which a first week may not reach.

Google's learning period page times learning in conversion events and conversion cycles. It says it can take up to around 50 conversion events or 3 conversion cycles for the bid strategy to calibrate to the new objective, although it can be faster depending on the amount of conversion data present.

Google's example of a conversion cycle is the amount of time it takes for a click to result in a conversion. On an account where people buy the same day, that wait is short. On one where they take a fortnight to decide, it is not, and a Friday verdict on a Monday change may be reading a window the strategy has not finished.

Meta says an ad set usually exits the learning phase after about 50 results in the week after its last significant edit. It publishes what counts as one: any change to targeting, to ad creative or to the optimisation event, adding a new ad to the ad set, pausing the ad set for seven days or longer, and changing bid strategy. It adds that a change to the budget amount, or to the bid control, cost per result goal or ROAS goal amount, may or may not be significant, depending on the magnitude of the change.

It also tells you when an ad set is unlikely to finish learning. Meta says an ad set becomes learning limited when it is unlikely to receive about 50 optimisation events in the week after your last significant edit. Meta treats that as a problem with the setup, and every fix it lists is a change to the setup.

This is our own practice, not either platform's. Before we call a bid change good or bad, we answer four questions in writing. What was the average target over the window, not the target today. How many conversion cycles have passed since the change. On Meta, when was the last significant edit, and did anything since reset it. And is the ad set able to finish learning at all, or is it learning limited.

If any of those answers is unknown, the honest verdict is not yet. That is less satisfying than a chart on a Friday, and it is the difference between a decision and a reaction.